GradCAM_YOLOv7

• Everyone knows the YOLO series model always can get good performance on detection tasks.
• I want to know does YOLO really learns the feature information or not.
• I will find time to add the code description

YOLO series model

• I already prepared and published the technique description on the internet. [HERE]
• One-stage methods prioritize inference speed, and models include YOLO, SSD, and RetinaNet.
• You can also refer to this website I wrote for how to train the YOLO model. [HERE]

GradCAM

• Explainable AI for computer vision. This can be used for diagnosing model predictions, either in production or while developing models.
• Weight the 2D activations by the average gradient.

YOLOv7 origin code needs to modify

1. ./model/yolo.py
2. ./model/experimental.py

Additional codes

1. ./model/gradcam.py
2. ./model/yolov7_object_detector.py
3. main_gradcam.py

Results